Title: Retrieval of snow cover area by optical sensors Rune Solberg, NR Contributions from: Petra Malcher a
1Retrieval of snow cover area by optical sensors
Rune Solberg, NRContributions from Petra
Malcher and Helmuth Rott, IMGISari Mätsimäki,
SYKE Miia Eskelinen and Martti Hallikainen, HUT
2Background
- Parameter Snow Covered Area (SCA) for
mountainous regions and the boreal forest zone - Definition Snow/no snow or areal fraction of
ground covered by snow - State of the art Daily information can be
retrieved from optical sensors giving data of 250
m 1 km spatial resolution. Visible,
near-infrared and thermal infrared part of the
electromagnetic spectrum - Motivation
- Snow cover is included in most hydrological
models - Important for water management and hydropower
- Important for flood prediction
- Important for energy balance calculations
- Problems
- Clouds
- Dynamic reflectance properties of snow
- Dense forest
- Steep terrain
3EnviSnow development The mountains
- Developed and validated a binary (snow/no snow)
SCA algorithms for MODIS and MERIS in the Alps - Developed and validated a new fractional SCA
algorithm for MODIS in the Norwegian mountains
4MODIS Snow Classification
- automated binary classification
- constrained to quality controlled data
- confined to cloud free pixels
- lake map (250 m) prevents water bodies to be
masked as snow -
- Cloud
- CO2 threshold test
- BT35(13.9 mm) lt 236 K
- BT31-BT22 test
- BT31(11 mm ) - BT22(3.9 mm) lt -12 K
- Seasonal BT31 threshold test
- BT31(11 mm) lt 240 K 01, 240 K 02 ...
- Band 1 6 reflectance tests
- Bd6(1.64 mm)/Bd1(0.65 mm) gt 0.6
- Bd1(0.65 mm) gt 20
- Snow
- NDSI of bands 3 (0.469 mm) and 6 (1.640 mm)
- NDSI gt 0.4
- BT31 thermal mask
- BT31(11 mm ) lt 282 K
- Bd3 threshold
- Bd3 gt 6
22 April 2005, MODIS 250 m
5ASTER/MODIS Snow Cover Difference
- Difference map for 20 June 2003, wider Ötztal
area - Overestimates broken snow cover
- Underestimates isolated snow patches mainly along
the snow rim - Experience with forested areas
- Overall better performance for non-forested areas
- Percentage of SCA mapped by both sensors
decreases with advancing snowmelt - Progressive fragmentation of snow cover is the
prime factor for the increase of snow fraction
mapped by ASTER only
only MODIS SCA
only ASTER SCA
20 June 2003, Ötztal
6MERIS Snow Classification
- semi-automated binary algorithm
- constrained to quality controlled data
- clouds are manually selected
- confined to cloud free pixels
- main snow classifier is the multi- temporal
ratio of band 3 (0.413 mm) - lake map (250 m) prevents water bodies to be
masked as snow
snow (others)
snow (forest)
clouds
snow free
water
bad input
27 April 2004
7Fractional SCA
- Typical problems with current operational
algorithms - Terrain effects
- Snow metamorphosis
- Anisotropic reflectance
- Snow impurities
- New algorithm
- Combines empirical and physical models for
compensation of the effects
8Fractional SCA algorithm
- Prior SCA estimation
- Prediction of
- Metamorphosis
- Impurities
- Linear spectral unmixing
- Iterative
9Experimental results
- Landsat TM and Terra MODIS images acquired 30 May
2004 - Late snowmelt season situation with very patchy
snow - Errors typically reduced from 40 in steep
slopes to lt5 - Errors for snow with high content of impurities
seen to be reduced from 15-20 (underestimation)
to lt5
10EnviSnow development Boreal forests
- SCA algorithms for boreal forests expanded from
AVHRR application to Terra MODIS and Envisat
MERIS - Operational dual-sensor SCA mapping using both
AVHRR and MODIS best image is chosen (based on
imaging geometry and cloud cover) - Airborne spectrometer measurements conducted for
determining snow reflectance properties in the
melting season
11(No Transcript)
12(No Transcript)
13Serious floods in Lapland May23 May30
14Airborne spectrometer campaigns
- Airborne measurements conducted for studying snow
reflectance properties in the melting season - Tuusula Test Site in latitude 60 N
- Sodankyla-Lokka Test Site in latitude 67 N
- Objective to assist satellite based snow cover
mapping algorithm development - Statistical variability for various land surface
reflectances in the boreal forest belt - Study the varying viewing angle effect on the
remote sensing reflectances with airborne
spectrometer
15Results
(1)
(3)
(1)Thick forest (2)Sparse forest (3) Melting snow
(2)
(4) Snow
- The AISA airborne spectrometer data used for
calculating statistics for snow and forest
reflectances has a pixel size of 1 m x 1 m
- Here is a visualised example of the averaged AISA
sample spectra collection for thick forest (1),
sparse forest (2) and melting snow (3)
16(a)
(b)
(c)
(d)
- Spectra from the AISA dataset acquired on 27
March 2002 in Tuusula. Averaged sample spectra
for (a) tree crowns in dense forest, (b) shadow
in dense forest, (c) dense forest and (d) sparse
forest
17Spectra from the AISA dataset acquired on 4 May
2003 in Sodankyla-Lokka test area
- The effect of variable observation angle to the
airborne spectrometer-derived radiance is
demonstrated for (a) principal solar plane and
(b) perpendicular plane - The land cover types are () snow-covered dense
forest, (o) snow-covered open area, (?)
snow-covered bog and (x) snow free ground in
wavelengths 551-555 nm
18Conclusions
- SCA algorithms for the mountains
- Binary SCA algorithms developed and validated for
MODIS and MERIS in the Alps (IMGI) - Fractional (at sub-pixel level) SCA algorithm
developed and tested for MODIS in the Norwegian
mountains tests indicate that it is
significantly more accurate than current
operationally used algorithm (NR) - SCA algorithms for boreal forest
- Fractional (at basin level) SCA algorithm
developed for boreal forest and validated for
AVHRR, MODIS and MERIS in Finland (SYKE) - Dual-sensor algorithm (AVHRR MODIS) in
operational use (SYKE) - Reflectance of snow and forest at varying density
measured in Finland in the melting season to
assist algorithm development (HUT)